net/ipv4/fib_semantics.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   4  *              operating system.  INET is implemented using the  BSD Socket
   5  *              interface as the means of communication with the user level.
   6  *
   7  *              IPv4 Forwarding Information Base: semantics.
   8  *
   9  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  10  */
  11
  12 #include <linux/uaccess.h>
  13 #include <linux/bitops.h>
  14 #include <linux/types.h>
  15 #include <linux/kernel.h>
  16 #include <linux/jiffies.h>
  17 #include <linux/mm.h>
  18 #include <linux/string.h>
  19 #include <linux/socket.h>
  20 #include <linux/sockios.h>
  21 #include <linux/errno.h>
  22 #include <linux/in.h>
  23 #include <linux/inet.h>
  24 #include <linux/inetdevice.h>
  25 #include <linux/netdevice.h>
  26 #include <linux/if_arp.h>
  27 #include <linux/proc_fs.h>
  28 #include <linux/skbuff.h>
  29 #include <linux/init.h>
  30 #include <linux/slab.h>
  31 #include <linux/netlink.h>
  32
  33 #include <net/arp.h>
  34 #include <net/ip.h>
  35 #include <net/protocol.h>
  36 #include <net/route.h>
  37 #include <net/tcp.h>
  38 #include <net/sock.h>
  39 #include <net/ip_fib.h>
  40 #include <net/ip6_fib.h>
  41 #include <net/nexthop.h>
  42 #include <net/netlink.h>
  43 #include <net/rtnh.h>
  44 #include <net/lwtunnel.h>
  45 #include <net/fib_notifier.h>
  46 #include <net/addrconf.h>
  47
  48 #include "fib_lookup.h"
  49
  50 static DEFINE_SPINLOCK(fib_info_lock);
  51 static struct hlist_head *fib_info_hash;
  52 static struct hlist_head *fib_info_laddrhash;
  53 static unsigned int fib_info_hash_size;
  54 static unsigned int fib_info_cnt;
  55
  56 #define DEVINDEX_HASHBITS 8
  57 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
  58 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
  59
  60 /* for_nexthops and change_nexthops only used when nexthop object
  61  * is not set in a fib_info. The logic within can reference fib_nh.
  62  */
  63 #ifdef CONFIG_IP_ROUTE_MULTIPATH
  64
  65 #define for_nexthops(fi) {                                              \
  66         int nhsel; const struct fib_nh *nh;                             \
  67         for (nhsel = 0, nh = (fi)->fib_nh;                              \
  68              nhsel < fib_info_num_path((fi));                           \
  69              nh++, nhsel++)
  70
  71 #define change_nexthops(fi) {                                           \
  72         int nhsel; struct fib_nh *nexthop_nh;                           \
  73         for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh);   \
  74              nhsel < fib_info_num_path((fi));                           \
  75              nexthop_nh++, nhsel++)
  76
  77 #else /* CONFIG_IP_ROUTE_MULTIPATH */
  78
  79 /* Hope, that gcc will optimize it to get rid of dummy loop */
  80
  81 #define for_nexthops(fi) {                                              \
  82         int nhsel; const struct fib_nh *nh = (fi)->fib_nh;              \
  83         for (nhsel = 0; nhsel < 1; nhsel++)
  84
  85 #define change_nexthops(fi) {                                           \
  86         int nhsel;                                                      \
  87         struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh);    \
  88         for (nhsel = 0; nhsel < 1; nhsel++)
  89
  90 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
  91
  92 #define endfor_nexthops(fi) }
  93
  94
  95 const struct fib_prop fib_props[RTN_MAX + 1] = {
  96         [RTN_UNSPEC] = {
  97                 .error  = 0,
  98                 .scope  = RT_SCOPE_NOWHERE,
  99         },
 100         [RTN_UNICAST] = {
 101                 .error  = 0,
 102                 .scope  = RT_SCOPE_UNIVERSE,
 103         },
 104         [RTN_LOCAL] = {
 105                 .error  = 0,
 106                 .scope  = RT_SCOPE_HOST,
 107         },
 108         [RTN_BROADCAST] = {
 109                 .error  = 0,
 110                 .scope  = RT_SCOPE_LINK,
 111         },
 112         [RTN_ANYCAST] = {
 113                 .error  = 0,
 114                 .scope  = RT_SCOPE_LINK,
 115         },
 116         [RTN_MULTICAST] = {
 117                 .error  = 0,
 118                 .scope  = RT_SCOPE_UNIVERSE,
 119         },
 120         [RTN_BLACKHOLE] = {
 121                 .error  = -EINVAL,
 122                 .scope  = RT_SCOPE_UNIVERSE,
 123         },
 124         [RTN_UNREACHABLE] = {
 125                 .error  = -EHOSTUNREACH,
 126                 .scope  = RT_SCOPE_UNIVERSE,
 127         },
 128         [RTN_PROHIBIT] = {
 129                 .error  = -EACCES,
 130                 .scope  = RT_SCOPE_UNIVERSE,
 131         },
 132         [RTN_THROW] = {
 133                 .error  = -EAGAIN,
 134                 .scope  = RT_SCOPE_UNIVERSE,
 135         },
 136         [RTN_NAT] = {
 137                 .error  = -EINVAL,
 138                 .scope  = RT_SCOPE_NOWHERE,
 139         },
 140         [RTN_XRESOLVE] = {
 141                 .error  = -EINVAL,
 142                 .scope  = RT_SCOPE_NOWHERE,
 143         },
 144 };
 145
 146 static void rt_fibinfo_free(struct rtable __rcu **rtp)
 147 {
 148         struct rtable *rt = rcu_dereference_protected(*rtp, 1);
 149
 150         if (!rt)
 151                 return;
 152
 153         /* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
 154          * because we waited an RCU grace period before calling
 155          * free_fib_info_rcu()
 156          */
 157
 158         dst_dev_put(&rt->dst);
 159         dst_release_immediate(&rt->dst);
 160 }
 161
 162 static void free_nh_exceptions(struct fib_nh_common *nhc)
 163 {
 164         struct fnhe_hash_bucket *hash;
 165         int i;
 166
 167         hash = rcu_dereference_protected(nhc->nhc_exceptions, 1);
 168         if (!hash)
 169                 return;
 170         for (i = 0; i < FNHE_HASH_SIZE; i++) {
 171                 struct fib_nh_exception *fnhe;
 172
 173                 fnhe = rcu_dereference_protected(hash[i].chain, 1);
 174                 while (fnhe) {
 175                         struct fib_nh_exception *next;
 176
 177                         next = rcu_dereference_protected(fnhe->fnhe_next, 1);
 178
 179                         rt_fibinfo_free(&fnhe->fnhe_rth_input);
 180                         rt_fibinfo_free(&fnhe->fnhe_rth_output);
 181
 182                         kfree(fnhe);
 183
 184                         fnhe = next;
 185                 }
 186         }
 187         kfree(hash);
 188 }
 189
 190 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
 191 {
 192         int cpu;
 193
 194         if (!rtp)
 195                 return;
 196
 197         for_each_possible_cpu(cpu) {
 198                 struct rtable *rt;
 199
 200                 rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
 201                 if (rt) {
 202                         dst_dev_put(&rt->dst);
 203                         dst_release_immediate(&rt->dst);
 204                 }
 205         }
 206         free_percpu(rtp);
 207 }
 208
 209 void fib_nh_common_release(struct fib_nh_common *nhc)
 210 {
 211         if (nhc->nhc_dev)
 212                 dev_put(nhc->nhc_dev);
 213
 214         lwtstate_put(nhc->nhc_lwtstate);
 215         rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
 216         rt_fibinfo_free(&nhc->nhc_rth_input);
 217         free_nh_exceptions(nhc);
 218 }
 219 EXPORT_SYMBOL_GPL(fib_nh_common_release);
 220
 221 void fib_nh_release(struct net *net, struct fib_nh *fib_nh)
 222 {
 223 #ifdef CONFIG_IP_ROUTE_CLASSID
 224         if (fib_nh->nh_tclassid)
 225                 net->ipv4.fib_num_tclassid_users--;
 226 #endif
 227         fib_nh_common_release(&fib_nh->nh_common);
 228 }
 229
 230 /* Release a nexthop info record */
 231 static void free_fib_info_rcu(struct rcu_head *head)
 232 {
 233         struct fib_info *fi = container_of(head, struct fib_info, rcu);
 234
 235         if (fi->nh) {
 236                 nexthop_put(fi->nh);
 237         } else {
 238                 change_nexthops(fi) {
 239                         fib_nh_release(fi->fib_net, nexthop_nh);
 240                 } endfor_nexthops(fi);
 241         }
 242
 243         ip_fib_metrics_put(fi->fib_metrics);
 244
 245         kfree(fi);
 246 }
 247
 248 void free_fib_info(struct fib_info *fi)
 249 {
 250         if (fi->fib_dead == 0) {
 251                 pr_warn("Freeing alive fib_info %p\n", fi);
 252                 return;
 253         }
 254         fib_info_cnt--;
 255
 256         call_rcu(&fi->rcu, free_fib_info_rcu);
 257 }
 258 EXPORT_SYMBOL_GPL(free_fib_info);
 259
 260 void fib_release_info(struct fib_info *fi)
 261 {
 262         spin_lock_bh(&fib_info_lock);
 263         if (fi && --fi->fib_treeref == 0) {
 264                 hlist_del(&fi->fib_hash);
 265                 if (fi->fib_prefsrc)
 266                         hlist_del(&fi->fib_lhash);
 267                 if (fi->nh) {
 268                         list_del(&fi->nh_list);
 269                 } else {
 270                         change_nexthops(fi) {
 271                                 if (!nexthop_nh->fib_nh_dev)
 272                                         continue;
 273                                 hlist_del(&nexthop_nh->nh_hash);
 274                         } endfor_nexthops(fi)
 275                 }
 276                 fi->fib_dead = 1;
 277                 fib_info_put(fi);
 278         }
 279         spin_unlock_bh(&fib_info_lock);
 280 }
 281
 282 static inline int nh_comp(struct fib_info *fi, struct fib_info *ofi)
 283 {
 284         const struct fib_nh *onh;
 285
 286         if (fi->nh || ofi->nh)
 287                 return nexthop_cmp(fi->nh, ofi->nh) ? 0 : -1;
 288
 289         if (ofi->fib_nhs == 0)
 290                 return 0;
 291
 292         for_nexthops(fi) {
 293                 onh = fib_info_nh(ofi, nhsel);
 294
 295                 if (nh->fib_nh_oif != onh->fib_nh_oif ||
 296                     nh->fib_nh_gw_family != onh->fib_nh_gw_family ||
 297                     nh->fib_nh_scope != onh->fib_nh_scope ||
 298 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 299                     nh->fib_nh_weight != onh->fib_nh_weight ||
 300 #endif
 301 #ifdef CONFIG_IP_ROUTE_CLASSID
 302                     nh->nh_tclassid != onh->nh_tclassid ||
 303 #endif
 304                     lwtunnel_cmp_encap(nh->fib_nh_lws, onh->fib_nh_lws) ||
 305                     ((nh->fib_nh_flags ^ onh->fib_nh_flags) & ~RTNH_COMPARE_MASK))
 306                         return -1;
 307
 308                 if (nh->fib_nh_gw_family == AF_INET &&
 309                     nh->fib_nh_gw4 != onh->fib_nh_gw4)
 310                         return -1;
 311
 312                 if (nh->fib_nh_gw_family == AF_INET6 &&
 313                     ipv6_addr_cmp(&nh->fib_nh_gw6, &onh->fib_nh_gw6))
 314                         return -1;
 315         } endfor_nexthops(fi);
 316         return 0;
 317 }
 318
 319 static inline unsigned int fib_devindex_hashfn(unsigned int val)
 320 {
 321         unsigned int mask = DEVINDEX_HASHSIZE - 1;
 322
 323         return (val ^
 324                 (val >> DEVINDEX_HASHBITS) ^
 325                 (val >> (DEVINDEX_HASHBITS * 2))) & mask;
 326 }
 327
 328 static unsigned int fib_info_hashfn_1(int init_val, u8 protocol, u8 scope,
 329                                       u32 prefsrc, u32 priority)
 330 {
 331         unsigned int val = init_val;
 332
 333         val ^= (protocol << 8) | scope;
 334         val ^= prefsrc;
 335         val ^= priority;
 336
 337         return val;
 338 }
 339
 340 static unsigned int fib_info_hashfn_result(unsigned int val)
 341 {
 342         unsigned int mask = (fib_info_hash_size - 1);
 343
 344         return (val ^ (val >> 7) ^ (val >> 12)) & mask;
 345 }
 346
 347 static inline unsigned int fib_info_hashfn(struct fib_info *fi)
 348 {
 349         unsigned int val;
 350
 351         val = fib_info_hashfn_1(fi->fib_nhs, fi->fib_protocol,
 352                                 fi->fib_scope, (__force u32)fi->fib_prefsrc,
 353                                 fi->fib_priority);
 354
 355         if (fi->nh) {
 356                 val ^= fib_devindex_hashfn(fi->nh->id);
 357         } else {
 358                 for_nexthops(fi) {
 359                         val ^= fib_devindex_hashfn(nh->fib_nh_oif);
 360                 } endfor_nexthops(fi)
 361         }
 362
 363         return fib_info_hashfn_result(val);
 364 }
 365
 366 /* no metrics, only nexthop id */
 367 static struct fib_info *fib_find_info_nh(struct net *net,
 368                                          const struct fib_config *cfg)
 369 {
 370         struct hlist_head *head;
 371         struct fib_info *fi;
 372         unsigned int hash;
 373
 374         hash = fib_info_hashfn_1(fib_devindex_hashfn(cfg->fc_nh_id),
 375                                  cfg->fc_protocol, cfg->fc_scope,
 376                                  (__force u32)cfg->fc_prefsrc,
 377                                  cfg->fc_priority);
 378         hash = fib_info_hashfn_result(hash);
 379         head = &fib_info_hash[hash];
 380
 381         hlist_for_each_entry(fi, head, fib_hash) {
 382                 if (!net_eq(fi->fib_net, net))
 383                         continue;
 384                 if (!fi->nh || fi->nh->id != cfg->fc_nh_id)
 385                         continue;
 386                 if (cfg->fc_protocol == fi->fib_protocol &&
 387                     cfg->fc_scope == fi->fib_scope &&
 388                     cfg->fc_prefsrc == fi->fib_prefsrc &&
 389                     cfg->fc_priority == fi->fib_priority &&
 390                     cfg->fc_type == fi->fib_type &&
 391                     cfg->fc_table == fi->fib_tb_id &&
 392                     !((cfg->fc_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK))
 393                         return fi;
 394         }
 395
 396         return NULL;
 397 }
 398
 399 static struct fib_info *fib_find_info(struct fib_info *nfi)
 400 {
 401         struct hlist_head *head;
 402         struct fib_info *fi;
 403         unsigned int hash;
 404
 405         hash = fib_info_hashfn(nfi);
 406         head = &fib_info_hash[hash];
 407
 408         hlist_for_each_entry(fi, head, fib_hash) {
 409                 if (!net_eq(fi->fib_net, nfi->fib_net))
 410                         continue;
 411                 if (fi->fib_nhs != nfi->fib_nhs)
 412                         continue;
 413                 if (nfi->fib_protocol == fi->fib_protocol &&
 414                     nfi->fib_scope == fi->fib_scope &&
 415                     nfi->fib_prefsrc == fi->fib_prefsrc &&
 416                     nfi->fib_priority == fi->fib_priority &&
 417                     nfi->fib_type == fi->fib_type &&
 418                     memcmp(nfi->fib_metrics, fi->fib_metrics,
 419                            sizeof(u32) * RTAX_MAX) == 0 &&
 420                     !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
 421                     nh_comp(fi, nfi) == 0)
 422                         return fi;
 423         }
 424
 425         return NULL;
 426 }
 427
 428 /* Check, that the gateway is already configured.
 429  * Used only by redirect accept routine.
 430  */
 431 int ip_fib_check_default(__be32 gw, struct net_device *dev)
 432 {
 433         struct hlist_head *head;
 434         struct fib_nh *nh;
 435         unsigned int hash;
 436
 437         spin_lock(&fib_info_lock);
 438
 439         hash = fib_devindex_hashfn(dev->ifindex);
 440         head = &fib_info_devhash[hash];
 441         hlist_for_each_entry(nh, head, nh_hash) {
 442                 if (nh->fib_nh_dev == dev &&
 443                     nh->fib_nh_gw4 == gw &&
 444                     !(nh->fib_nh_flags & RTNH_F_DEAD)) {
 445                         spin_unlock(&fib_info_lock);
 446                         return 0;
 447                 }
 448         }
 449
 450         spin_unlock(&fib_info_lock);
 451
 452         return -1;
 453 }
 454
 455 static inline size_t fib_nlmsg_size(struct fib_info *fi)
 456 {
 457         size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
 458                          + nla_total_size(4) /* RTA_TABLE */
 459                          + nla_total_size(4) /* RTA_DST */
 460                          + nla_total_size(4) /* RTA_PRIORITY */
 461                          + nla_total_size(4) /* RTA_PREFSRC */
 462                          + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
 463         unsigned int nhs = fib_info_num_path(fi);
 464
 465         /* space for nested metrics */
 466         payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
 467
 468         if (fi->nh)
 469                 payload += nla_total_size(4); /* RTA_NH_ID */
 470
 471         if (nhs) {
 472                 size_t nh_encapsize = 0;
 473                 /* Also handles the special case nhs == 1 */
 474
 475                 /* each nexthop is packed in an attribute */
 476                 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
 477                 unsigned int i;
 478
 479                 /* may contain flow and gateway attribute */
 480                 nhsize += 2 * nla_total_size(4);
 481
 482                 /* grab encap info */
 483                 for (i = 0; i < fib_info_num_path(fi); i++) {
 484                         struct fib_nh_common *nhc = fib_info_nhc(fi, i);
 485
 486                         if (nhc->nhc_lwtstate) {
 487                                 /* RTA_ENCAP_TYPE */
 488                                 nh_encapsize += lwtunnel_get_encap_size(
 489                                                 nhc->nhc_lwtstate);
 490                                 /* RTA_ENCAP */
 491                                 nh_encapsize +=  nla_total_size(2);
 492                         }
 493                 }
 494
 495                 /* all nexthops are packed in a nested attribute */
 496                 payload += nla_total_size((nhs * nhsize) + nh_encapsize);
 497
 498         }
 499
 500         return payload;
 501 }
 502
 503 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
 504                int dst_len, u32 tb_id, const struct nl_info *info,
 505                unsigned int nlm_flags)
 506 {
 507         struct fib_rt_info fri;
 508         struct sk_buff *skb;
 509         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
 510         int err = -ENOBUFS;
 511
 512         skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
 513         if (!skb)
 514                 goto errout;
 515
 516         fri.fi = fa->fa_info;
 517         fri.tb_id = tb_id;
 518         fri.dst = key;
 519         fri.dst_len = dst_len;
 520         fri.tos = fa->fa_tos;
 521         fri.type = fa->fa_type;
 522         err = fib_dump_info(skb, info->portid, seq, event, &fri, nlm_flags);
 523         if (err < 0) {
 524                 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
 525                 WARN_ON(err == -EMSGSIZE);
 526                 kfree_skb(skb);
 527                 goto errout;
 528         }
 529         rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
 530                     info->nlh, GFP_KERNEL);
 531         return;
 532 errout:
 533         if (err < 0)
 534                 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
 535 }
 536
 537 static int fib_detect_death(struct fib_info *fi, int order,
 538                             struct fib_info **last_resort, int *last_idx,
 539                             int dflt)
 540 {
 541         const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
 542         struct neighbour *n;
 543         int state = NUD_NONE;
 544
 545         if (likely(nhc->nhc_gw_family == AF_INET))
 546                 n = neigh_lookup(&arp_tbl, &nhc->nhc_gw.ipv4, nhc->nhc_dev);
 547         else if (nhc->nhc_gw_family == AF_INET6)
 548                 n = neigh_lookup(ipv6_stub->nd_tbl, &nhc->nhc_gw.ipv6,
 549                                  nhc->nhc_dev);
 550         else
 551                 n = NULL;
 552
 553         if (n) {
 554                 state = n->nud_state;
 555                 neigh_release(n);
 556         } else {
 557                 return 0;
 558         }
 559         if (state == NUD_REACHABLE)
 560                 return 0;
 561         if ((state & NUD_VALID) && order != dflt)
 562                 return 0;
 563         if ((state & NUD_VALID) ||
 564             (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
 565                 *last_resort = fi;
 566                 *last_idx = order;
 567         }
 568         return 1;
 569 }
 570
 571 int fib_nh_common_init(struct fib_nh_common *nhc, struct nlattr *encap,
 572                        u16 encap_type, void *cfg, gfp_t gfp_flags,
 573                        struct netlink_ext_ack *extack)
 574 {
 575         int err;
 576
 577         nhc->nhc_pcpu_rth_output = alloc_percpu_gfp(struct rtable __rcu *,
 578                                                     gfp_flags);
 579         if (!nhc->nhc_pcpu_rth_output)
 580                 return -ENOMEM;
 581
 582         if (encap) {
 583                 struct lwtunnel_state *lwtstate;
 584
 585                 if (encap_type == LWTUNNEL_ENCAP_NONE) {
 586                         NL_SET_ERR_MSG(extack, "LWT encap type not specified");
 587                         err = -EINVAL;
 588                         goto lwt_failure;
 589                 }
 590                 err = lwtunnel_build_state(encap_type, encap, nhc->nhc_family,
 591                                            cfg, &lwtstate, extack);
 592                 if (err)
 593                         goto lwt_failure;
 594
 595                 nhc->nhc_lwtstate = lwtstate_get(lwtstate);
 596         }
 597
 598         return 0;
 599
 600 lwt_failure:
 601         rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
 602         nhc->nhc_pcpu_rth_output = NULL;
 603         return err;
 604 }
 605 EXPORT_SYMBOL_GPL(fib_nh_common_init);
 606
 607 int fib_nh_init(struct net *net, struct fib_nh *nh,
 608                 struct fib_config *cfg, int nh_weight,
 609                 struct netlink_ext_ack *extack)
 610 {
 611         int err;
 612
 613         nh->fib_nh_family = AF_INET;
 614
 615         err = fib_nh_common_init(&nh->nh_common, cfg->fc_encap,
 616                                  cfg->fc_encap_type, cfg, GFP_KERNEL, extack);
 617         if (err)
 618                 return err;
 619
 620         nh->fib_nh_oif = cfg->fc_oif;
 621         nh->fib_nh_gw_family = cfg->fc_gw_family;
 622         if (cfg->fc_gw_family == AF_INET)
 623                 nh->fib_nh_gw4 = cfg->fc_gw4;
 624         else if (cfg->fc_gw_family == AF_INET6)
 625                 nh->fib_nh_gw6 = cfg->fc_gw6;
 626
 627         nh->fib_nh_flags = cfg->fc_flags;
 628
 629 #ifdef CONFIG_IP_ROUTE_CLASSID
 630         nh->nh_tclassid = cfg->fc_flow;
 631         if (nh->nh_tclassid)
 632                 net->ipv4.fib_num_tclassid_users++;
 633 #endif
 634 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 635         nh->fib_nh_weight = nh_weight;
 636 #endif
 637         return 0;
 638 }
 639
 640 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 641
 642 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
 643                               struct netlink_ext_ack *extack)
 644 {
 645         int nhs = 0;
 646
 647         while (rtnh_ok(rtnh, remaining)) {
 648                 nhs++;
 649                 rtnh = rtnh_next(rtnh, &remaining);
 650         }
 651
 652         /* leftover implies invalid nexthop configuration, discard it */
 653         if (remaining > 0) {
 654                 NL_SET_ERR_MSG(extack,
 655                                "Invalid nexthop configuration - extra data after nexthops");
 656                 nhs = 0;
 657         }
 658
 659         return nhs;
 660 }
 661
 662 /* only called when fib_nh is integrated into fib_info */
 663 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
 664                        int remaining, struct fib_config *cfg,
 665                        struct netlink_ext_ack *extack)
 666 {
 667         struct net *net = fi->fib_net;
 668         struct fib_config fib_cfg;
 669         struct fib_nh *nh;
 670         int ret;
 671
 672         change_nexthops(fi) {
 673                 int attrlen;
 674
 675                 memset(&fib_cfg, 0, sizeof(fib_cfg));
 676
 677                 if (!rtnh_ok(rtnh, remaining)) {
 678                         NL_SET_ERR_MSG(extack,
 679                                        "Invalid nexthop configuration - extra data after nexthop");
 680                         return -EINVAL;
 681                 }
 682
 683                 if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
 684                         NL_SET_ERR_MSG(extack,
 685                                        "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
 686                         return -EINVAL;
 687                 }
 688
 689                 fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
 690                 fib_cfg.fc_oif = rtnh->rtnh_ifindex;
 691
 692                 attrlen = rtnh_attrlen(rtnh);
 693                 if (attrlen > 0) {
 694                         struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
 695
 696                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 697                         nlav = nla_find(attrs, attrlen, RTA_VIA);
 698                         if (nla && nlav) {
 699                                 NL_SET_ERR_MSG(extack,
 700                                                "Nexthop configuration can not contain both GATEWAY and VIA");
 701                                 return -EINVAL;
 702                         }
 703                         if (nla) {
 704                                 fib_cfg.fc_gw4 = nla_get_in_addr(nla);
 705                                 if (fib_cfg.fc_gw4)
 706                                         fib_cfg.fc_gw_family = AF_INET;
 707                         } else if (nlav) {
 708                                 ret = fib_gw_from_via(&fib_cfg, nlav, extack);
 709                                 if (ret)
 710                                         goto errout;
 711                         }
 712
 713                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 714                         if (nla)
 715                                 fib_cfg.fc_flow = nla_get_u32(nla);
 716
 717                         fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
 718                         nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
 719                         if (nla)
 720                                 fib_cfg.fc_encap_type = nla_get_u16(nla);
 721                 }
 722
 723                 ret = fib_nh_init(net, nexthop_nh, &fib_cfg,
 724                                   rtnh->rtnh_hops + 1, extack);
 725                 if (ret)
 726                         goto errout;
 727
 728                 rtnh = rtnh_next(rtnh, &remaining);
 729         } endfor_nexthops(fi);
 730
 731         ret = -EINVAL;
 732         nh = fib_info_nh(fi, 0);
 733         if (cfg->fc_oif && nh->fib_nh_oif != cfg->fc_oif) {
 734                 NL_SET_ERR_MSG(extack,
 735                                "Nexthop device index does not match RTA_OIF");
 736                 goto errout;
 737         }
 738         if (cfg->fc_gw_family) {
 739                 if (cfg->fc_gw_family != nh->fib_nh_gw_family ||
 740                     (cfg->fc_gw_family == AF_INET &&
 741                      nh->fib_nh_gw4 != cfg->fc_gw4) ||
 742                     (cfg->fc_gw_family == AF_INET6 &&
 743                      ipv6_addr_cmp(&nh->fib_nh_gw6, &cfg->fc_gw6))) {
 744                         NL_SET_ERR_MSG(extack,
 745                                        "Nexthop gateway does not match RTA_GATEWAY or RTA_VIA");
 746                         goto errout;
 747                 }
 748         }
 749 #ifdef CONFIG_IP_ROUTE_CLASSID
 750         if (cfg->fc_flow && nh->nh_tclassid != cfg->fc_flow) {
 751                 NL_SET_ERR_MSG(extack,
 752                                "Nexthop class id does not match RTA_FLOW");
 753                 goto errout;
 754         }
 755 #endif
 756         ret = 0;
 757 errout:
 758         return ret;
 759 }
 760
 761 /* only called when fib_nh is integrated into fib_info */
 762 static void fib_rebalance(struct fib_info *fi)
 763 {
 764         int total;
 765         int w;
 766
 767         if (fib_info_num_path(fi) < 2)
 768                 return;
 769
 770         total = 0;
 771         for_nexthops(fi) {
 772                 if (nh->fib_nh_flags & RTNH_F_DEAD)
 773                         continue;
 774
 775                 if (ip_ignore_linkdown(nh->fib_nh_dev) &&
 776                     nh->fib_nh_flags & RTNH_F_LINKDOWN)
 777                         continue;
 778
 779                 total += nh->fib_nh_weight;
 780         } endfor_nexthops(fi);
 781
 782         w = 0;
 783         change_nexthops(fi) {
 784                 int upper_bound;
 785
 786                 if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) {
 787                         upper_bound = -1;
 788                 } else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) &&
 789                            nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
 790                         upper_bound = -1;
 791                 } else {
 792                         w += nexthop_nh->fib_nh_weight;
 793                         upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
 794                                                             total) - 1;
 795                 }
 796
 797                 atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound);
 798         } endfor_nexthops(fi);
 799 }
 800 #else /* CONFIG_IP_ROUTE_MULTIPATH */
 801
 802 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
 803                        int remaining, struct fib_config *cfg,
 804                        struct netlink_ext_ack *extack)
 805 {
 806         NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel");
 807
 808         return -EINVAL;
 809 }
 810
 811 #define fib_rebalance(fi) do { } while (0)
 812
 813 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
 814
 815 static int fib_encap_match(u16 encap_type,
 816                            struct nlattr *encap,
 817                            const struct fib_nh *nh,
 818                            const struct fib_config *cfg,
 819                            struct netlink_ext_ack *extack)
 820 {
 821         struct lwtunnel_state *lwtstate;
 822         int ret, result = 0;
 823
 824         if (encap_type == LWTUNNEL_ENCAP_NONE)
 825                 return 0;
 826
 827         ret = lwtunnel_build_state(encap_type, encap, AF_INET,
 828                                    cfg, &lwtstate, extack);
 829         if (!ret) {
 830                 result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws);
 831                 lwtstate_free(lwtstate);
 832         }
 833
 834         return result;
 835 }
 836
 837 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi,
 838                  struct netlink_ext_ack *extack)
 839 {
 840 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 841         struct rtnexthop *rtnh;
 842         int remaining;
 843 #endif
 844
 845         if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
 846                 return 1;
 847
 848         if (cfg->fc_nh_id) {
 849                 if (fi->nh && cfg->fc_nh_id == fi->nh->id)
 850                         return 0;
 851                 return 1;
 852         }
 853
 854         if (cfg->fc_oif || cfg->fc_gw_family) {
 855                 struct fib_nh *nh = fib_info_nh(fi, 0);
 856
 857                 if (cfg->fc_encap) {
 858                         if (fib_encap_match(cfg->fc_encap_type, cfg->fc_encap,
 859                                             nh, cfg, extack))
 860                                 return 1;
 861                 }
 862 #ifdef CONFIG_IP_ROUTE_CLASSID
 863                 if (cfg->fc_flow &&
 864                     cfg->fc_flow != nh->nh_tclassid)
 865                         return 1;
 866 #endif
 867                 if ((cfg->fc_oif && cfg->fc_oif != nh->fib_nh_oif) ||
 868                     (cfg->fc_gw_family &&
 869                      cfg->fc_gw_family != nh->fib_nh_gw_family))
 870                         return 1;
 871
 872                 if (cfg->fc_gw_family == AF_INET &&
 873                     cfg->fc_gw4 != nh->fib_nh_gw4)
 874                         return 1;
 875
 876                 if (cfg->fc_gw_family == AF_INET6 &&
 877                     ipv6_addr_cmp(&cfg->fc_gw6, &nh->fib_nh_gw6))
 878                         return 1;
 879
 880                 return 0;
 881         }
 882
 883 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 884         if (!cfg->fc_mp)
 885                 return 0;
 886
 887         rtnh = cfg->fc_mp;
 888         remaining = cfg->fc_mp_len;
 889
 890         for_nexthops(fi) {
 891                 int attrlen;
 892
 893                 if (!rtnh_ok(rtnh, remaining))
 894                         return -EINVAL;
 895
 896                 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif)
 897                         return 1;
 898
 899                 attrlen = rtnh_attrlen(rtnh);
 900                 if (attrlen > 0) {
 901                         struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
 902
 903                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 904                         nlav = nla_find(attrs, attrlen, RTA_VIA);
 905                         if (nla && nlav) {
 906                                 NL_SET_ERR_MSG(extack,
 907                                                "Nexthop configuration can not contain both GATEWAY and VIA");
 908                                 return -EINVAL;
 909                         }
 910
 911                         if (nla) {
 912                                 if (nh->fib_nh_gw_family != AF_INET ||
 913                                     nla_get_in_addr(nla) != nh->fib_nh_gw4)
 914                                         return 1;
 915                         } else if (nlav) {
 916                                 struct fib_config cfg2;
 917                                 int err;
 918
 919                                 err = fib_gw_from_via(&cfg2, nlav, extack);
 920                                 if (err)
 921                                         return err;
 922
 923                                 switch (nh->fib_nh_gw_family) {
 924                                 case AF_INET:
 925                                         if (cfg2.fc_gw_family != AF_INET ||
 926                                             cfg2.fc_gw4 != nh->fib_nh_gw4)
 927                                                 return 1;
 928                                         break;
 929                                 case AF_INET6:
 930                                         if (cfg2.fc_gw_family != AF_INET6 ||
 931                                             ipv6_addr_cmp(&cfg2.fc_gw6,
 932                                                           &nh->fib_nh_gw6))
 933                                                 return 1;
 934                                         break;
 935                                 }
 936                         }
 937
 938 #ifdef CONFIG_IP_ROUTE_CLASSID
 939                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 940                         if (nla && nla_get_u32(nla) != nh->nh_tclassid)
 941                                 return 1;
 942 #endif
 943                 }
 944
 945                 rtnh = rtnh_next(rtnh, &remaining);
 946         } endfor_nexthops(fi);
 947 #endif
 948         return 0;
 949 }
 950
 951 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
 952 {
 953         struct nlattr *nla;
 954         int remaining;
 955
 956         if (!cfg->fc_mx)
 957                 return true;
 958
 959         nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
 960                 int type = nla_type(nla);
 961                 u32 fi_val, val;
 962
 963                 if (!type)
 964                         continue;
 965                 if (type > RTAX_MAX)
 966                         return false;
 967
 968                 if (type == RTAX_CC_ALGO) {
 969                         char tmp[TCP_CA_NAME_MAX];
 970                         bool ecn_ca = false;
 971
 972                         nla_strlcpy(tmp, nla, sizeof(tmp));
 973                         val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
 974                 } else {
 975                         if (nla_len(nla) != sizeof(u32))
 976                                 return false;
 977                         val = nla_get_u32(nla);
 978                 }
 979
 980                 fi_val = fi->fib_metrics->metrics[type - 1];
 981                 if (type == RTAX_FEATURES)
 982                         fi_val &= ~DST_FEATURE_ECN_CA;
 983
 984                 if (fi_val != val)
 985                         return false;
 986         }
 987
 988         return true;
 989 }
 990
 991 static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh,
 992                               u32 table, struct netlink_ext_ack *extack)
 993 {
 994         struct fib6_config cfg = {
 995                 .fc_table = table,
 996                 .fc_flags = nh->fib_nh_flags | RTF_GATEWAY,
 997                 .fc_ifindex = nh->fib_nh_oif,
 998                 .fc_gateway = nh->fib_nh_gw6,
 999         };
1000         struct fib6_nh fib6_nh = {};
1001         int err;
1002
1003         err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack);
1004         if (!err) {
1005                 nh->fib_nh_dev = fib6_nh.fib_nh_dev;
1006                 dev_hold(nh->fib_nh_dev);
1007                 nh->fib_nh_oif = nh->fib_nh_dev->ifindex;
1008                 nh->fib_nh_scope = RT_SCOPE_LINK;
1009
1010                 ipv6_stub->fib6_nh_release(&fib6_nh);
1011         }
1012
1013         return err;
1014 }
1015
1016 /*
1017  * Picture
1018  * -------
1019  *
1020  * Semantics of nexthop is very messy by historical reasons.
1021  * We have to take into account, that:
1022  * a) gateway can be actually local interface address,
1023  *    so that gatewayed route is direct.
1024  * b) gateway must be on-link address, possibly
1025  *    described not by an ifaddr, but also by a direct route.
1026  * c) If both gateway and interface are specified, they should not
1027  *    contradict.
1028  * d) If we use tunnel routes, gateway could be not on-link.
1029  *
1030  * Attempt to reconcile all of these (alas, self-contradictory) conditions
1031  * results in pretty ugly and hairy code with obscure logic.
1032  *
1033  * I chose to generalized it instead, so that the size
1034  * of code does not increase practically, but it becomes
1035  * much more general.
1036  * Every prefix is assigned a "scope" value: "host" is local address,
1037  * "link" is direct route,
1038  * [ ... "site" ... "interior" ... ]
1039  * and "universe" is true gateway route with global meaning.
1040  *
1041  * Every prefix refers to a set of "nexthop"s (gw, oif),
1042  * where gw must have narrower scope. This recursion stops
1043  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
1044  * which means that gw is forced to be on link.
1045  *
1046  * Code is still hairy, but now it is apparently logically
1047  * consistent and very flexible. F.e. as by-product it allows
1048  * to co-exists in peace independent exterior and interior
1049  * routing processes.
1050  *
1051  * Normally it looks as following.
1052  *
1053  * {universe prefix}  -> (gw, oif) [scope link]
1054  *                |
1055  *                |-> {link prefix} -> (gw, oif) [scope local]
1056  *                                      |
1057  *                                      |-> {local prefix} (terminal node)
1058  */
1059 static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table,
1060                               u8 scope, struct netlink_ext_ack *extack)
1061 {
1062         struct net_device *dev;
1063         struct fib_result res;
1064         int err = 0;
1065
1066         if (nh->fib_nh_flags & RTNH_F_ONLINK) {
1067                 unsigned int addr_type;
1068
1069                 if (scope >= RT_SCOPE_LINK) {
1070                         NL_SET_ERR_MSG(extack, "Nexthop has invalid scope");
1071                         return -EINVAL;
1072                 }
1073                 dev = __dev_get_by_index(net, nh->fib_nh_oif);
1074                 if (!dev) {
1075                         NL_SET_ERR_MSG(extack, "Nexthop device required for onlink");
1076                         return -ENODEV;
1077                 }
1078                 if (!(dev->flags & IFF_UP)) {
1079                         NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1080                         return -ENETDOWN;
1081                 }
1082                 addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4);
1083                 if (addr_type != RTN_UNICAST) {
1084                         NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1085                         return -EINVAL;
1086                 }
1087                 if (!netif_carrier_ok(dev))
1088                         nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1089                 nh->fib_nh_dev = dev;
1090                 dev_hold(dev);
1091                 nh->fib_nh_scope = RT_SCOPE_LINK;
1092                 return 0;
1093         }
1094         rcu_read_lock();
1095         {
1096                 struct fib_table *tbl = NULL;
1097                 struct flowi4 fl4 = {
1098                         .daddr = nh->fib_nh_gw4,
1099                         .flowi4_scope = scope + 1,
1100                         .flowi4_oif = nh->fib_nh_oif,
1101                         .flowi4_iif = LOOPBACK_IFINDEX,
1102                 };
1103
1104                 /* It is not necessary, but requires a bit of thinking */
1105                 if (fl4.flowi4_scope < RT_SCOPE_LINK)
1106                         fl4.flowi4_scope = RT_SCOPE_LINK;
1107
1108                 if (table)
1109                         tbl = fib_get_table(net, table);
1110
1111                 if (tbl)
1112                         err = fib_table_lookup(tbl, &fl4, &res,
1113                                                FIB_LOOKUP_IGNORE_LINKSTATE |
1114                                                FIB_LOOKUP_NOREF);
1115
1116                 /* on error or if no table given do full lookup. This
1117                  * is needed for example when nexthops are in the local
1118                  * table rather than the given table
1119                  */
1120                 if (!tbl || err) {
1121                         err = fib_lookup(net, &fl4, &res,
1122                                          FIB_LOOKUP_IGNORE_LINKSTATE);
1123                 }
1124
1125                 if (err) {
1126                         NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1127                         goto out;
1128                 }
1129         }
1130
1131         err = -EINVAL;
1132         if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
1133                 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1134                 goto out;
1135         }
1136         nh->fib_nh_scope = res.scope;
1137         nh->fib_nh_oif = FIB_RES_OIF(res);
1138         nh->fib_nh_dev = dev = FIB_RES_DEV(res);
1139         if (!dev) {
1140                 NL_SET_ERR_MSG(extack,
1141                                "No egress device for nexthop gateway");
1142                 goto out;
1143         }
1144         dev_hold(dev);
1145         if (!netif_carrier_ok(dev))
1146                 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1147         err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
1148 out:
1149         rcu_read_unlock();
1150         return err;
1151 }
1152
1153 static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh,
1154                               struct netlink_ext_ack *extack)
1155 {
1156         struct in_device *in_dev;
1157         int err;
1158
1159         if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
1160                 NL_SET_ERR_MSG(extack,
1161                                "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
1162                 return -EINVAL;
1163         }
1164
1165         rcu_read_lock();
1166
1167         err = -ENODEV;
1168         in_dev = inetdev_by_index(net, nh->fib_nh_oif);
1169         if (!in_dev)
1170                 goto out;
1171         err = -ENETDOWN;
1172         if (!(in_dev->dev->flags & IFF_UP)) {
1173                 NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
1174                 goto out;
1175         }
1176
1177         nh->fib_nh_dev = in_dev->dev;
1178         dev_hold(nh->fib_nh_dev);
1179         nh->fib_nh_scope = RT_SCOPE_HOST;
1180         if (!netif_carrier_ok(nh->fib_nh_dev))
1181                 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1182         err = 0;
1183 out:
1184         rcu_read_unlock();
1185         return err;
1186 }
1187
1188 int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
1189                  struct netlink_ext_ack *extack)
1190 {
1191         int err;
1192
1193         if (nh->fib_nh_gw_family == AF_INET)
1194                 err = fib_check_nh_v4_gw(net, nh, table, scope, extack);
1195         else if (nh->fib_nh_gw_family == AF_INET6)
1196                 err = fib_check_nh_v6_gw(net, nh, table, extack);
1197         else
1198                 err = fib_check_nh_nongw(net, nh, extack);
1199
1200         return err;
1201 }
1202
1203 static inline unsigned int fib_laddr_hashfn(__be32 val)
1204 {
1205         unsigned int mask = (fib_info_hash_size - 1);
1206
1207         return ((__force u32)val ^
1208                 ((__force u32)val >> 7) ^
1209                 ((__force u32)val >> 14)) & mask;
1210 }
1211
1212 static struct hlist_head *fib_info_hash_alloc(int bytes)
1213 {
1214         if (bytes <= PAGE_SIZE)
1215                 return kzalloc(bytes, GFP_KERNEL);
1216         else
1217                 return (struct hlist_head *)
1218                         __get_free_pages(GFP_KERNEL | __GFP_ZERO,
1219                                          get_order(bytes));
1220 }
1221
1222 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
1223 {
1224         if (!hash)
1225                 return;
1226
1227         if (bytes <= PAGE_SIZE)
1228                 kfree(hash);
1229         else
1230                 free_pages((unsigned long) hash, get_order(bytes));
1231 }
1232
1233 static void fib_info_hash_move(struct hlist_head *new_info_hash,
1234                                struct hlist_head *new_laddrhash,
1235                                unsigned int new_size)
1236 {
1237         struct hlist_head *old_info_hash, *old_laddrhash;
1238         unsigned int old_size = fib_info_hash_size;
1239         unsigned int i, bytes;
1240
1241         spin_lock_bh(&fib_info_lock);
1242         old_info_hash = fib_info_hash;
1243         old_laddrhash = fib_info_laddrhash;
1244         fib_info_hash_size = new_size;
1245
1246         for (i = 0; i < old_size; i++) {
1247                 struct hlist_head *head = &fib_info_hash[i];
1248                 struct hlist_node *n;
1249                 struct fib_info *fi;
1250
1251                 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
1252                         struct hlist_head *dest;
1253                         unsigned int new_hash;
1254
1255                         new_hash = fib_info_hashfn(fi);
1256                         dest = &new_info_hash[new_hash];
1257                         hlist_add_head(&fi->fib_hash, dest);
1258                 }
1259         }
1260         fib_info_hash = new_info_hash;
1261
1262         for (i = 0; i < old_size; i++) {
1263                 struct hlist_head *lhead = &fib_info_laddrhash[i];
1264                 struct hlist_node *n;
1265                 struct fib_info *fi;
1266
1267                 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
1268                         struct hlist_head *ldest;
1269                         unsigned int new_hash;
1270
1271                         new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
1272                         ldest = &new_laddrhash[new_hash];
1273                         hlist_add_head(&fi->fib_lhash, ldest);
1274                 }
1275         }
1276         fib_info_laddrhash = new_laddrhash;
1277
1278         spin_unlock_bh(&fib_info_lock);
1279
1280         bytes = old_size * sizeof(struct hlist_head *);
1281         fib_info_hash_free(old_info_hash, bytes);
1282         fib_info_hash_free(old_laddrhash, bytes);
1283 }
1284
1285 __be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc,
1286                                  unsigned char scope)
1287 {
1288         struct fib_nh *nh;
1289
1290         if (nhc->nhc_family != AF_INET)
1291                 return inet_select_addr(nhc->nhc_dev, 0, scope);
1292
1293         nh = container_of(nhc, struct fib_nh, nh_common);
1294         nh->nh_saddr = inet_select_addr(nh->fib_nh_dev, nh->fib_nh_gw4, scope);
1295         nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
1296
1297         return nh->nh_saddr;
1298 }
1299
1300 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res)
1301 {
1302         struct fib_nh_common *nhc = res->nhc;
1303
1304         if (res->fi->fib_prefsrc)
1305                 return res->fi->fib_prefsrc;
1306
1307         if (nhc->nhc_family == AF_INET) {
1308                 struct fib_nh *nh;
1309
1310                 nh = container_of(nhc, struct fib_nh, nh_common);
1311                 if (nh->nh_saddr_genid == atomic_read(&net->ipv4.dev_addr_genid))
1312                         return nh->nh_saddr;
1313         }
1314
1315         return fib_info_update_nhc_saddr(net, nhc, res->fi->fib_scope);
1316 }
1317
1318 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
1319 {
1320         if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1321             fib_prefsrc != cfg->fc_dst) {
1322                 u32 tb_id = cfg->fc_table;
1323                 int rc;
1324
1325                 if (tb_id == RT_TABLE_MAIN)
1326                         tb_id = RT_TABLE_LOCAL;
1327
1328                 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1329                                           fib_prefsrc, tb_id);
1330
1331                 if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1332                         rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1333                                                   fib_prefsrc, RT_TABLE_LOCAL);
1334                 }
1335
1336                 if (rc != RTN_LOCAL)
1337                         return false;
1338         }
1339         return true;
1340 }
1341
1342 struct fib_info *fib_create_info(struct fib_config *cfg,
1343                                  struct netlink_ext_ack *extack)
1344 {
1345         int err;
1346         struct fib_info *fi = NULL;
1347         struct nexthop *nh = NULL;
1348         struct fib_info *ofi;
1349         int nhs = 1;
1350         struct net *net = cfg->fc_nlinfo.nl_net;
1351
1352         if (cfg->fc_type > RTN_MAX)
1353                 goto err_inval;
1354
1355         /* Fast check to catch the most weird cases */
1356         if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1357                 NL_SET_ERR_MSG(extack, "Invalid scope");
1358                 goto err_inval;
1359         }
1360
1361         if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1362                 NL_SET_ERR_MSG(extack,
1363                                "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1364                 goto err_inval;
1365         }
1366
1367         if (cfg->fc_nh_id) {
1368                 if (!cfg->fc_mx) {
1369                         fi = fib_find_info_nh(net, cfg);
1370                         if (fi) {
1371                                 fi->fib_treeref++;
1372                                 return fi;
1373                         }
1374                 }
1375
1376                 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
1377                 if (!nh) {
1378                         NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
1379                         goto err_inval;
1380                 }
1381                 nhs = 0;
1382         }
1383
1384 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1385         if (cfg->fc_mp) {
1386                 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1387                 if (nhs == 0)
1388                         goto err_inval;
1389         }
1390 #endif
1391
1392         err = -ENOBUFS;
1393         if (fib_info_cnt >= fib_info_hash_size) {
1394                 unsigned int new_size = fib_info_hash_size << 1;
1395                 struct hlist_head *new_info_hash;
1396                 struct hlist_head *new_laddrhash;
1397                 unsigned int bytes;
1398
1399                 if (!new_size)
1400                         new_size = 16;
1401                 bytes = new_size * sizeof(struct hlist_head *);
1402                 new_info_hash = fib_info_hash_alloc(bytes);
1403                 new_laddrhash = fib_info_hash_alloc(bytes);
1404                 if (!new_info_hash || !new_laddrhash) {
1405                         fib_info_hash_free(new_info_hash, bytes);
1406                         fib_info_hash_free(new_laddrhash, bytes);
1407                 } else
1408                         fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1409
1410                 if (!fib_info_hash_size)
1411                         goto failure;
1412         }
1413
1414         fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL);
1415         if (!fi)
1416                 goto failure;
1417         fi->fib_metrics = ip_fib_metrics_init(fi->fib_net, cfg->fc_mx,
1418                                               cfg->fc_mx_len, extack);
1419         if (IS_ERR(fi->fib_metrics)) {
1420                 err = PTR_ERR(fi->fib_metrics);
1421                 kfree(fi);
1422                 return ERR_PTR(err);
1423         }
1424
1425         fib_info_cnt++;
1426         fi->fib_net = net;
1427         fi->fib_protocol = cfg->fc_protocol;
1428         fi->fib_scope = cfg->fc_scope;
1429         fi->fib_flags = cfg->fc_flags;
1430         fi->fib_priority = cfg->fc_priority;
1431         fi->fib_prefsrc = cfg->fc_prefsrc;
1432         fi->fib_type = cfg->fc_type;
1433         fi->fib_tb_id = cfg->fc_table;
1434
1435         fi->fib_nhs = nhs;
1436         if (nh) {
1437                 if (!nexthop_get(nh)) {
1438                         NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
1439                         err = -EINVAL;
1440                 } else {
1441                         err = 0;
1442                         fi->nh = nh;
1443                 }
1444         } else {
1445                 change_nexthops(fi) {
1446                         nexthop_nh->nh_parent = fi;
1447                 } endfor_nexthops(fi)
1448
1449                 if (cfg->fc_mp)
1450                         err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg,
1451                                           extack);
1452                 else
1453                         err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack);
1454         }
1455
1456         if (err != 0)
1457                 goto failure;
1458
1459         if (fib_props[cfg->fc_type].error) {
1460                 if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) {
1461                         NL_SET_ERR_MSG(extack,
1462                                        "Gateway, device and multipath can not be specified for this route type");
1463                         goto err_inval;
1464                 }
1465                 goto link_it;
1466         } else {
1467                 switch (cfg->fc_type) {
1468                 case RTN_UNICAST:
1469                 case RTN_LOCAL:
1470                 case RTN_BROADCAST:
1471                 case RTN_ANYCAST:
1472                 case RTN_MULTICAST:
1473                         break;
1474                 default:
1475                         NL_SET_ERR_MSG(extack, "Invalid route type");
1476                         goto err_inval;
1477                 }
1478         }
1479
1480         if (cfg->fc_scope > RT_SCOPE_HOST) {
1481                 NL_SET_ERR_MSG(extack, "Invalid scope");
1482                 goto err_inval;
1483         }
1484
1485         if (fi->nh) {
1486                 err = fib_check_nexthop(fi->nh, cfg->fc_scope, extack);
1487                 if (err)
1488                         goto failure;
1489         } else if (cfg->fc_scope == RT_SCOPE_HOST) {
1490                 struct fib_nh *nh = fi->fib_nh;
1491
1492                 /* Local address is added. */
1493                 if (nhs != 1) {
1494                         NL_SET_ERR_MSG(extack,
1495                                        "Route with host scope can not have multiple nexthops");
1496                         goto err_inval;
1497                 }
1498                 if (nh->fib_nh_gw_family) {
1499                         NL_SET_ERR_MSG(extack,
1500                                        "Route with host scope can not have a gateway");
1501                         goto err_inval;
1502                 }
1503                 nh->fib_nh_scope = RT_SCOPE_NOWHERE;
1504                 nh->fib_nh_dev = dev_get_by_index(net, nh->fib_nh_oif);
1505                 err = -ENODEV;
1506                 if (!nh->fib_nh_dev)
1507                         goto failure;
1508         } else {
1509                 int linkdown = 0;
1510
1511                 change_nexthops(fi) {
1512                         err = fib_check_nh(cfg->fc_nlinfo.nl_net, nexthop_nh,
1513                                            cfg->fc_table, cfg->fc_scope,
1514                                            extack);
1515                         if (err != 0)
1516                                 goto failure;
1517                         if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN)
1518                                 linkdown++;
1519                 } endfor_nexthops(fi)
1520                 if (linkdown == fi->fib_nhs)
1521                         fi->fib_flags |= RTNH_F_LINKDOWN;
1522         }
1523
1524         if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1525                 NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1526                 goto err_inval;
1527         }
1528
1529         if (!fi->nh) {
1530                 change_nexthops(fi) {
1531                         fib_info_update_nhc_saddr(net, &nexthop_nh->nh_common,
1532                                                   fi->fib_scope);
1533                         if (nexthop_nh->fib_nh_gw_family == AF_INET6)
1534                                 fi->fib_nh_is_v6 = true;
1535                 } endfor_nexthops(fi)
1536
1537                 fib_rebalance(fi);
1538         }
1539
1540 link_it:
1541         ofi = fib_find_info(fi);
1542         if (ofi) {
1543                 fi->fib_dead = 1;
1544                 free_fib_info(fi);
1545                 ofi->fib_treeref++;
1546                 return ofi;
1547         }
1548
1549         fi->fib_treeref++;
1550         refcount_set(&fi->fib_clntref, 1);
1551         spin_lock_bh(&fib_info_lock);
1552         hlist_add_head(&fi->fib_hash,
1553                        &fib_info_hash[fib_info_hashfn(fi)]);
1554         if (fi->fib_prefsrc) {
1555                 struct hlist_head *head;
1556
1557                 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1558                 hlist_add_head(&fi->fib_lhash, head);
1559         }
1560         if (fi->nh) {
1561                 list_add(&fi->nh_list, &nh->fi_list);
1562         } else {
1563                 change_nexthops(fi) {
1564                         struct hlist_head *head;
1565                         unsigned int hash;
1566
1567                         if (!nexthop_nh->fib_nh_dev)
1568                                 continue;
1569                         hash = fib_devindex_hashfn(nexthop_nh->fib_nh_dev->ifindex);
1570                         head = &fib_info_devhash[hash];
1571                         hlist_add_head(&nexthop_nh->nh_hash, head);
1572                 } endfor_nexthops(fi)
1573         }
1574         spin_unlock_bh(&fib_info_lock);
1575         return fi;
1576
1577 err_inval:
1578         err = -EINVAL;
1579
1580 failure:
1581         if (fi) {
1582                 fi->fib_dead = 1;
1583                 free_fib_info(fi);
1584         }
1585
1586         return ERR_PTR(err);
1587 }
1588
1589 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc,
1590                      u8 rt_family, unsigned char *flags, bool skip_oif)
1591 {
1592         if (nhc->nhc_flags & RTNH_F_DEAD)
1593                 *flags |= RTNH_F_DEAD;
1594
1595         if (nhc->nhc_flags & RTNH_F_LINKDOWN) {
1596                 *flags |= RTNH_F_LINKDOWN;
1597
1598                 rcu_read_lock();
1599                 switch (nhc->nhc_family) {
1600                 case AF_INET:
1601                         if (ip_ignore_linkdown(nhc->nhc_dev))
1602                                 *flags |= RTNH_F_DEAD;
1603                         break;
1604                 case AF_INET6:
1605                         if (ip6_ignore_linkdown(nhc->nhc_dev))
1606                                 *flags |= RTNH_F_DEAD;
1607                         break;
1608                 }
1609                 rcu_read_unlock();
1610         }
1611
1612         switch (nhc->nhc_gw_family) {
1613         case AF_INET:
1614                 if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4))
1615                         goto nla_put_failure;
1616                 break;
1617         case AF_INET6:
1618                 /* if gateway family does not match nexthop family
1619                  * gateway is encoded as RTA_VIA
1620                  */
1621                 if (rt_family != nhc->nhc_gw_family) {
1622                         int alen = sizeof(struct in6_addr);
1623                         struct nlattr *nla;
1624                         struct rtvia *via;
1625
1626                         nla = nla_reserve(skb, RTA_VIA, alen + 2);
1627                         if (!nla)
1628                                 goto nla_put_failure;
1629
1630                         via = nla_data(nla);
1631                         via->rtvia_family = AF_INET6;
1632                         memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen);
1633                 } else if (nla_put_in6_addr(skb, RTA_GATEWAY,
1634                                             &nhc->nhc_gw.ipv6) < 0) {
1635                         goto nla_put_failure;
1636                 }
1637                 break;
1638         }
1639
1640         *flags |= (nhc->nhc_flags & RTNH_F_ONLINK);
1641         if (nhc->nhc_flags & RTNH_F_OFFLOAD)
1642                 *flags |= RTNH_F_OFFLOAD;
1643
1644         if (!skip_oif && nhc->nhc_dev &&
1645             nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex))
1646                 goto nla_put_failure;
1647
1648         if (nhc->nhc_lwtstate &&
1649             lwtunnel_fill_encap(skb, nhc->nhc_lwtstate,
1650                                 RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
1651                 goto nla_put_failure;
1652
1653         return 0;
1654
1655 nla_put_failure:
1656         return -EMSGSIZE;
1657 }
1658 EXPORT_SYMBOL_GPL(fib_nexthop_info);
1659
1660 #if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6)
1661 int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc,
1662                     int nh_weight, u8 rt_family)
1663 {
1664         const struct net_device *dev = nhc->nhc_dev;
1665         struct rtnexthop *rtnh;
1666         unsigned char flags = 0;
1667
1668         rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1669         if (!rtnh)
1670                 goto nla_put_failure;
1671
1672         rtnh->rtnh_hops = nh_weight - 1;
1673         rtnh->rtnh_ifindex = dev ? dev->ifindex : 0;
1674
1675         if (fib_nexthop_info(skb, nhc, rt_family, &flags, true) < 0)
1676                 goto nla_put_failure;
1677
1678         rtnh->rtnh_flags = flags;
1679
1680         /* length of rtnetlink header + attributes */
1681         rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1682
1683         return 0;
1684
1685 nla_put_failure:
1686         return -EMSGSIZE;
1687 }
1688 EXPORT_SYMBOL_GPL(fib_add_nexthop);
1689 #endif
1690
1691 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1692 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1693 {
1694         struct nlattr *mp;
1695
1696         mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
1697         if (!mp)
1698                 goto nla_put_failure;
1699
1700         if (unlikely(fi->nh)) {
1701                 if (nexthop_mpath_fill_node(skb, fi->nh, AF_INET) < 0)
1702                         goto nla_put_failure;
1703                 goto mp_end;
1704         }
1705
1706         for_nexthops(fi) {
1707                 if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight,
1708                                     AF_INET) < 0)
1709                         goto nla_put_failure;
1710 #ifdef CONFIG_IP_ROUTE_CLASSID
1711                 if (nh->nh_tclassid &&
1712                     nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1713                         goto nla_put_failure;
1714 #endif
1715         } endfor_nexthops(fi);
1716
1717 mp_end:
1718         nla_nest_end(skb, mp);
1719
1720         return 0;
1721
1722 nla_put_failure:
1723         return -EMSGSIZE;
1724 }
1725 #else
1726 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1727 {
1728         return 0;
1729 }
1730 #endif
1731
1732 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1733                   struct fib_rt_info *fri, unsigned int flags)
1734 {
1735         unsigned int nhs = fib_info_num_path(fri->fi);
1736         struct fib_info *fi = fri->fi;
1737         u32 tb_id = fri->tb_id;
1738         struct nlmsghdr *nlh;
1739         struct rtmsg *rtm;
1740
1741         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1742         if (!nlh)
1743                 return -EMSGSIZE;
1744
1745         rtm = nlmsg_data(nlh);
1746         rtm->rtm_family = AF_INET;
1747         rtm->rtm_dst_len = fri->dst_len;
1748         rtm->rtm_src_len = 0;
1749         rtm->rtm_tos = fri->tos;
1750         if (tb_id < 256)
1751                 rtm->rtm_table = tb_id;
1752         else
1753                 rtm->rtm_table = RT_TABLE_COMPAT;
1754         if (nla_put_u32(skb, RTA_TABLE, tb_id))
1755                 goto nla_put_failure;
1756         rtm->rtm_type = fri->type;
1757         rtm->rtm_flags = fi->fib_flags;
1758         rtm->rtm_scope = fi->fib_scope;
1759         rtm->rtm_protocol = fi->fib_protocol;
1760
1761         if (rtm->rtm_dst_len &&
1762             nla_put_in_addr(skb, RTA_DST, fri->dst))
1763                 goto nla_put_failure;
1764         if (fi->fib_priority &&
1765             nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1766                 goto nla_put_failure;
1767         if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1768                 goto nla_put_failure;
1769
1770         if (fi->fib_prefsrc &&
1771             nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1772                 goto nla_put_failure;
1773
1774         if (fi->nh) {
1775                 if (nla_put_u32(skb, RTA_NH_ID, fi->nh->id))
1776                         goto nla_put_failure;
1777                 if (nexthop_is_blackhole(fi->nh))
1778                         rtm->rtm_type = RTN_BLACKHOLE;
1779         }
1780
1781         if (nhs == 1) {
1782                 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
1783                 unsigned char flags = 0;
1784
1785                 if (fib_nexthop_info(skb, nhc, AF_INET, &flags, false) < 0)
1786                         goto nla_put_failure;
1787
1788                 rtm->rtm_flags = flags;
1789 #ifdef CONFIG_IP_ROUTE_CLASSID
1790                 if (nhc->nhc_family == AF_INET) {
1791                         struct fib_nh *nh;
1792
1793                         nh = container_of(nhc, struct fib_nh, nh_common);
1794                         if (nh->nh_tclassid &&
1795                             nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1796                                 goto nla_put_failure;
1797                 }
1798 #endif
1799         } else {
1800                 if (fib_add_multipath(skb, fi) < 0)
1801                         goto nla_put_failure;
1802         }
1803
1804         nlmsg_end(skb, nlh);
1805         return 0;
1806
1807 nla_put_failure:
1808         nlmsg_cancel(skb, nlh);
1809         return -EMSGSIZE;
1810 }
1811
1812 /*
1813  * Update FIB if:
1814  * - local address disappeared -> we must delete all the entries
1815  *   referring to it.
1816  * - device went down -> we must shutdown all nexthops going via it.
1817  */
1818 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1819 {
1820         int ret = 0;
1821         unsigned int hash = fib_laddr_hashfn(local);
1822         struct hlist_head *head = &fib_info_laddrhash[hash];
1823         int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
1824         struct net *net = dev_net(dev);
1825         struct fib_info *fi;
1826
1827         if (!fib_info_laddrhash || local == 0)
1828                 return 0;
1829
1830         hlist_for_each_entry(fi, head, fib_lhash) {
1831                 if (!net_eq(fi->fib_net, net) ||
1832                     fi->fib_tb_id != tb_id)
1833                         continue;
1834                 if (fi->fib_prefsrc == local) {
1835                         fi->fib_flags |= RTNH_F_DEAD;
1836                         ret++;
1837                 }
1838         }
1839         return ret;
1840 }
1841
1842 static int call_fib_nh_notifiers(struct fib_nh *nh,
1843                                  enum fib_event_type event_type)
1844 {
1845         bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev);
1846         struct fib_nh_notifier_info info = {
1847                 .fib_nh = nh,
1848         };
1849
1850         switch (event_type) {
1851         case FIB_EVENT_NH_ADD:
1852                 if (nh->fib_nh_flags & RTNH_F_DEAD)
1853                         break;
1854                 if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN)
1855                         break;
1856                 return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type,
1857                                            &info.info);
1858         case FIB_EVENT_NH_DEL:
1859                 if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) ||
1860                     (nh->fib_nh_flags & RTNH_F_DEAD))
1861                         return call_fib4_notifiers(dev_net(nh->fib_nh_dev),
1862                                                    event_type, &info.info);
1863         default:
1864                 break;
1865         }
1866
1867         return NOTIFY_DONE;
1868 }
1869
1870 /* Update the PMTU of exceptions when:
1871  * - the new MTU of the first hop becomes smaller than the PMTU
1872  * - the old MTU was the same as the PMTU, and it limited discovery of
1873  *   larger MTUs on the path. With that limit raised, we can now
1874  *   discover larger MTUs
1875  * A special case is locked exceptions, for which the PMTU is smaller
1876  * than the minimal accepted PMTU:
1877  * - if the new MTU is greater than the PMTU, don't make any change
1878  * - otherwise, unlock and set PMTU
1879  */
1880 void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig)
1881 {
1882         struct fnhe_hash_bucket *bucket;
1883         int i;
1884
1885         bucket = rcu_dereference_protected(nhc->nhc_exceptions, 1);
1886         if (!bucket)
1887                 return;
1888
1889         for (i = 0; i < FNHE_HASH_SIZE; i++) {
1890                 struct fib_nh_exception *fnhe;
1891
1892                 for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1893                      fnhe;
1894                      fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1895                         if (fnhe->fnhe_mtu_locked) {
1896                                 if (new <= fnhe->fnhe_pmtu) {
1897                                         fnhe->fnhe_pmtu = new;
1898                                         fnhe->fnhe_mtu_locked = false;
1899                                 }
1900                         } else if (new < fnhe->fnhe_pmtu ||
1901                                    orig == fnhe->fnhe_pmtu) {
1902                                 fnhe->fnhe_pmtu = new;
1903                         }
1904                 }
1905         }
1906 }
1907
1908 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1909 {
1910         unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1911         struct hlist_head *head = &fib_info_devhash[hash];
1912         struct fib_nh *nh;
1913
1914         hlist_for_each_entry(nh, head, nh_hash) {
1915                 if (nh->fib_nh_dev == dev)
1916                         fib_nhc_update_mtu(&nh->nh_common, dev->mtu, orig_mtu);
1917         }
1918 }
1919
1920 /* Event              force Flags           Description
1921  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1922  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1923  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1924  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1925  *
1926  * only used when fib_nh is built into fib_info
1927  */
1928 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1929 {
1930         int ret = 0;
1931         int scope = RT_SCOPE_NOWHERE;
1932         struct fib_info *prev_fi = NULL;
1933         unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1934         struct hlist_head *head = &fib_info_devhash[hash];
1935         struct fib_nh *nh;
1936
1937         if (force)
1938                 scope = -1;
1939
1940         hlist_for_each_entry(nh, head, nh_hash) {
1941                 struct fib_info *fi = nh->nh_parent;
1942                 int dead;
1943
1944                 BUG_ON(!fi->fib_nhs);
1945                 if (nh->fib_nh_dev != dev || fi == prev_fi)
1946                         continue;
1947                 prev_fi = fi;
1948                 dead = 0;
1949                 change_nexthops(fi) {
1950                         if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD)
1951                                 dead++;
1952                         else if (nexthop_nh->fib_nh_dev == dev &&
1953                                  nexthop_nh->fib_nh_scope != scope) {
1954                                 switch (event) {
1955                                 case NETDEV_DOWN:
1956                                 case NETDEV_UNREGISTER:
1957                                         nexthop_nh->fib_nh_flags |= RTNH_F_DEAD;
1958                                         /* fall through */
1959                                 case NETDEV_CHANGE:
1960                                         nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1961                                         break;
1962                                 }
1963                                 call_fib_nh_notifiers(nexthop_nh,
1964                                                       FIB_EVENT_NH_DEL);
1965                                 dead++;
1966                         }
1967 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1968                         if (event == NETDEV_UNREGISTER &&
1969                             nexthop_nh->fib_nh_dev == dev) {
1970                                 dead = fi->fib_nhs;
1971                                 break;
1972                         }
1973 #endif
1974                 } endfor_nexthops(fi)
1975                 if (dead == fi->fib_nhs) {
1976                         switch (event) {
1977                         case NETDEV_DOWN:
1978                         case NETDEV_UNREGISTER:
1979                                 fi->fib_flags |= RTNH_F_DEAD;
1980                                 /* fall through */
1981                         case NETDEV_CHANGE:
1982                                 fi->fib_flags |= RTNH_F_LINKDOWN;
1983                                 break;
1984                         }
1985                         ret++;
1986                 }
1987
1988                 fib_rebalance(fi);
1989         }
1990
1991         return ret;
1992 }
1993
1994 /* Must be invoked inside of an RCU protected region.  */
1995 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1996 {
1997         struct fib_info *fi = NULL, *last_resort = NULL;
1998         struct hlist_head *fa_head = res->fa_head;
1999         struct fib_table *tb = res->table;
2000         u8 slen = 32 - res->prefixlen;
2001         int order = -1, last_idx = -1;
2002         struct fib_alias *fa, *fa1 = NULL;
2003         u32 last_prio = res->fi->fib_priority;
2004         u8 last_tos = 0;
2005
2006         hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
2007                 struct fib_info *next_fi = fa->fa_info;
2008                 struct fib_nh *nh;
2009
2010                 if (fa->fa_slen != slen)
2011                         continue;
2012                 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
2013                         continue;
2014                 if (fa->tb_id != tb->tb_id)
2015                         continue;
2016                 if (next_fi->fib_priority > last_prio &&
2017                     fa->fa_tos == last_tos) {
2018                         if (last_tos)
2019                                 continue;
2020                         break;
2021                 }
2022                 if (next_fi->fib_flags & RTNH_F_DEAD)
2023                         continue;
2024                 last_tos = fa->fa_tos;
2025                 last_prio = next_fi->fib_priority;
2026
2027                 if (next_fi->fib_scope != res->scope ||
2028                     fa->fa_type != RTN_UNICAST)
2029                         continue;
2030
2031                 nh = fib_info_nh(next_fi, 0);
2032                 if (!nh->fib_nh_gw4 || nh->fib_nh_scope != RT_SCOPE_LINK)
2033                         continue;
2034
2035                 fib_alias_accessed(fa);
2036
2037                 if (!fi) {
2038                         if (next_fi != res->fi)
2039                                 break;
2040                         fa1 = fa;
2041                 } else if (!fib_detect_death(fi, order, &last_resort,
2042                                              &last_idx, fa1->fa_default)) {
2043                         fib_result_assign(res, fi);
2044                         fa1->fa_default = order;
2045                         goto out;
2046                 }
2047                 fi = next_fi;
2048                 order++;
2049         }
2050
2051         if (order <= 0 || !fi) {
2052                 if (fa1)
2053                         fa1->fa_default = -1;
2054                 goto out;
2055         }
2056
2057         if (!fib_detect_death(fi, order, &last_resort, &last_idx,
2058                               fa1->fa_default)) {
2059                 fib_result_assign(res, fi);
2060                 fa1->fa_default = order;
2061                 goto out;
2062         }
2063
2064         if (last_idx >= 0)
2065                 fib_result_assign(res, last_resort);
2066         fa1->fa_default = last_idx;
2067 out:
2068         return;
2069 }
2070
2071 /*
2072  * Dead device goes up. We wake up dead nexthops.
2073  * It takes sense only on multipath routes.
2074  *
2075  * only used when fib_nh is built into fib_info
2076  */
2077 int fib_sync_up(struct net_device *dev, unsigned char nh_flags)
2078 {
2079         struct fib_info *prev_fi;
2080         unsigned int hash;
2081         struct hlist_head *head;
2082         struct fib_nh *nh;
2083         int ret;
2084
2085         if (!(dev->flags & IFF_UP))
2086                 return 0;
2087
2088         if (nh_flags & RTNH_F_DEAD) {
2089                 unsigned int flags = dev_get_flags(dev);
2090
2091                 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
2092                         nh_flags |= RTNH_F_LINKDOWN;
2093         }
2094
2095         prev_fi = NULL;
2096         hash = fib_devindex_hashfn(dev->ifindex);
2097         head = &fib_info_devhash[hash];
2098         ret = 0;
2099
2100         hlist_for_each_entry(nh, head, nh_hash) {
2101                 struct fib_info *fi = nh->nh_parent;
2102                 int alive;
2103
2104                 BUG_ON(!fi->fib_nhs);
2105                 if (nh->fib_nh_dev != dev || fi == prev_fi)
2106                         continue;
2107
2108                 prev_fi = fi;
2109                 alive = 0;
2110                 change_nexthops(fi) {
2111                         if (!(nexthop_nh->fib_nh_flags & nh_flags)) {
2112                                 alive++;
2113                                 continue;
2114                         }
2115                         if (!nexthop_nh->fib_nh_dev ||
2116                             !(nexthop_nh->fib_nh_dev->flags & IFF_UP))
2117                                 continue;
2118                         if (nexthop_nh->fib_nh_dev != dev ||
2119                             !__in_dev_get_rtnl(dev))
2120                                 continue;
2121                         alive++;
2122                         nexthop_nh->fib_nh_flags &= ~nh_flags;
2123                         call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
2124                 } endfor_nexthops(fi)
2125
2126                 if (alive > 0) {
2127                         fi->fib_flags &= ~nh_flags;
2128                         ret++;
2129                 }
2130
2131                 fib_rebalance(fi);
2132         }
2133
2134         return ret;
2135 }
2136
2137 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2138 static bool fib_good_nh(const struct fib_nh *nh)
2139 {
2140         int state = NUD_REACHABLE;
2141
2142         if (nh->fib_nh_scope == RT_SCOPE_LINK) {
2143                 struct neighbour *n;
2144
2145                 rcu_read_lock_bh();
2146
2147                 if (likely(nh->fib_nh_gw_family == AF_INET))
2148                         n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
2149                                                    (__force u32)nh->fib_nh_gw4);
2150                 else if (nh->fib_nh_gw_family == AF_INET6)
2151                         n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev,
2152                                                            &nh->fib_nh_gw6);
2153                 else
2154                         n = NULL;
2155                 if (n)
2156                         state = n->nud_state;
2157
2158                 rcu_read_unlock_bh();
2159         }
2160
2161         return !!(state & NUD_VALID);
2162 }
2163
2164 void fib_select_multipath(struct fib_result *res, int hash)
2165 {
2166         struct fib_info *fi = res->fi;
2167         struct net *net = fi->fib_net;
2168         bool first = false;
2169
2170         if (unlikely(res->fi->nh)) {
2171                 nexthop_path_fib_result(res, hash);
2172                 return;
2173         }
2174
2175         change_nexthops(fi) {
2176                 if (net->ipv4.sysctl_fib_multipath_use_neigh) {
2177                         if (!fib_good_nh(nexthop_nh))
2178                                 continue;
2179                         if (!first) {
2180                                 res->nh_sel = nhsel;
2181                                 res->nhc = &nexthop_nh->nh_common;
2182                                 first = true;
2183                         }
2184                 }
2185
2186                 if (hash > atomic_read(&nexthop_nh->fib_nh_upper_bound))
2187                         continue;
2188
2189                 res->nh_sel = nhsel;
2190                 res->nhc = &nexthop_nh->nh_common;
2191                 return;
2192         } endfor_nexthops(fi);
2193 }
2194 #endif
2195
2196 void fib_select_path(struct net *net, struct fib_result *res,
2197                      struct flowi4 *fl4, const struct sk_buff *skb)
2198 {
2199         if (fl4->flowi4_oif && !(fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF))
2200                 goto check_saddr;
2201
2202 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2203         if (fib_info_num_path(res->fi) > 1) {
2204                 int h = fib_multipath_hash(net, fl4, skb, NULL);
2205
2206                 fib_select_multipath(res, h);
2207         }
2208         else
2209 #endif
2210         if (!res->prefixlen &&
2211             res->table->tb_num_default > 1 &&
2212             res->type == RTN_UNICAST)
2213                 fib_select_default(fl4, res);
2214
2215 check_saddr:
2216         if (!fl4->saddr)
2217                 fl4->saddr = fib_result_prefsrc(net, res);
2218 }